[openib-general] [PATCH 1 of 3] mad: large RMPP support, Round 2

[Hal Rosenstock]

On Sun, 2006-02-12 at 10:27, Jack Morgenstein wrote:
> Implement large RMPP support:

I applied this patch series as it addresses Sean's primary objection in
terms of walking the RMPP list (being O(n) rather than O(n**2)) and
addressing most of his other comments. It certainly makes things better
than they are now and Sean can make further changes next week. 

Sean also had a comment about single segment MADs/single SGE relative to
mad.c::ib_create_send_mad which I don't see addressed but he did say he
would be willing to accept it without this.

It is important to start to get feedback on this from the larger
community (especially those with large clusters) prior to pushing this
upstream.

Also, user_mad.c::copy_recv_mad appears to still be copying to a
temporary buffer but I think that was what coaelsce was already doing.

A nit is to factor out the common data_offset code in mad_rmpp.c and
user_mad.c.

-- Hal